Promotion gift customization design system based on theme activities

Abstract

The invention discloses a promotion gift customization design system based on theme activities, and relates to the technical field of gift customization design, and the system comprises a multi-source data collection module, an intelligent feature extraction and evaluation module, a data classification module, and a risk level analysis and response module. The behavior data of the user on each platform and channel are obtained, and the comprehensive portrait of the user is constructed by integrating multi-dimensional data, so that a solid foundation is laid for subsequent analysis. According to the invention, through multi-source data acquisition, feature extraction, data classification and risk analysis, the system accurately identifies real demands of the user, filters disguise behaviors, generates a more fit gift design, and improves user satisfaction. Meanwhile, abnormal data and risk indexes are dynamically processed, the influence of camouflage behaviors on the model is reduced, data quality is ensured, resource allocation is optimized, system stability is enhanced, and reliable support is provided for subsequent activity design.

Description

Technical Field

[0001] The present invention relates to the technical field of gift customization design, and in particular to a promotional gift customization design system based on theme activities. Background Art

[0002] The promotional gift customization design system based on theme activities is an intelligent platform designed to help companies quickly and accurately design and customize promotional gifts based on the theme of a specific event or festival. The system integrates a variety of design and customization functions, allowing users to choose the appropriate gift type, material, color, pattern, etc. according to the needs of the event and the characteristics of the audience. Through intelligent algorithms and big data analysis, the system can analyze user needs and generate gift design solutions that meet the characteristics of the theme. At the same time, the system also supports mass production and multi-channel distribution, making the supply chain management of gifts more efficient, and improving the design efficiency and delivery speed of customized gifts while ensuring quality.

[0003] In actual applications, the promotional gift customization design system based on theme activities usually integrates technologies such as artificial intelligence, 3D modeling and personalized recommendations to better meet the diverse needs of the market. Users can upload brand logos, select theme elements, etc. in the system, and the system will automatically generate multiple gift design sketches for selection. In addition, the system also supports users to preview the effects of customized gifts and even conduct virtual trials to quickly iterate and optimize the design. For enterprises, this system not only reduces design costs and time investment, but also enhances the brand image and customer satisfaction of the event, thereby adding unique added value to marketing activities. This gift customization system, with theme activities as the core, opens up the entire process from design to production, enabling enterprises to achieve high-quality customized promotional gift delivery in a short period of time.

[0004] The prior art has the following deficiencies:

[0005] In the customized design of promotional gifts based on theme activities, the system usually collects and integrates user behavior data through big data. However, some user behavior data may not truly reflect their actual needs, but rather "disguised behaviors" generated due to specific situations or in order to complete platform tasks, obtain points rewards, etc. For example, users may frequently click on multiple products to obtain points, but these clicks do not represent their true preferences. Such disguised behaviors will generate abnormal data, which can easily mislead the judgment of the algorithm and cause the system to generate design solutions that do not meet actual needs. If these disguised behaviors are not identified, the system may mistakenly believe that these behaviors reflect the real needs of users, resulting in serious deviations in analysis and design. For example, the system may generate a gift design that does not match the user's interests based on false clicks, and the promotional gifts finally launched fail to meet user expectations, leading to user disappointment and affecting the event effect and brand image.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0007] The purpose of the present invention is to provide a promotional gift customization design system based on theme activities. Through multi-source data collection, intelligent feature extraction and evaluation, data classification and risk level analysis modules, the system accurately identifies the real needs of users, separates disguised behaviors, generates gift design solutions that are more in line with the needs, and improves user satisfaction and brand image. At the same time, by dynamically processing abnormal data and evaluating abnormal risk indexes, the system reduces the impact of disguised behaviors on the model, ensures data quality, and optimizes resource allocation, thereby improving the long-term stability and efficiency of the overall system, providing reliable support for subsequent activity design, and solving the problems in the above-mentioned background technology.

[0008] In order to achieve the above-mentioned purpose, the present invention provides the following technical solutions: a promotional gift customization design system based on theme activities, including a multi-source data acquisition module, an intelligent feature extraction and evaluation module, a data classification module, and a risk level analysis and response module:

[0009] The multi-source data collection module first conducts comprehensive and multi-source data collection to obtain user behavior data on various platforms and channels. By integrating multi-dimensional data, it builds a comprehensive portrait of the user, laying a solid foundation for subsequent analysis.

[0010] The intelligent feature extraction and evaluation module extracts features from the acquired behavior data, inputs the extracted feature information into the pre-learned machine learning model, and intelligently evaluates the acquired user behavior data to predict the authenticity and preference of user behavior and identify disguised behavior in the behavior data;

[0011] The data classification module divides the acquired behavior data into normal behavior data and abnormal behavior data based on the evaluation results of the machine learning model;

[0012] The risk level analysis and response module conducts further comprehensive analysis on the identified abnormal behavior data, determines the risk level of the abnormal behavior data, and takes different response measures for different risk levels.

[0013] Preferably, the steps to conduct comprehensive and multi-source data collection, build a comprehensive portrait of the user, and lay a solid foundation for subsequent analysis are as follows:

[0014] First, identify all data sources and determine the scope of user behavior data to be collected;

[0015] After determining the data source, set the corresponding data collection frequency and sampling strategy;

[0016] Preprocess the collected data;

[0017] After cleaning and standardization, data from different platforms and channels are integrated to form a comprehensive portrait of the user.

[0018] Preferably, feature extraction is performed on the acquired behavioral data, and the extracted features include the contrast degree between the user's click, browsing behavior and purchasing behavior and the correlation between the user behavior and the platform reward mechanism. After analyzing the contrast degree between the extracted user's click, browsing behavior and purchasing behavior, a purchase conversion contradiction index is generated. After analyzing the correlation between the extracted user behavior and the platform reward mechanism, a task-oriented index is generated. The purchase conversion contradiction index and the task-oriented index are used to identify disguised behavior in the behavioral data.

[0019] Preferably, after obtaining the purchase conversion contradiction index and the task orientation index, the purchase conversion contradiction index and the task orientation index are input into a pre-learned machine learning model, a camouflage coefficient is generated by the machine learning model, and the obtained behavioral data is intelligently evaluated by the camouflage coefficient.

[0020] Preferably, the camouflage coefficient generated after analyzing the acquired user behavior data is compared and analyzed with a preset camouflage coefficient reference threshold, and the acquired behavior data is divided, and the division steps are as follows:

[0021] If the camouflage coefficient is greater than or equal to a preset camouflage coefficient reference threshold, the acquired behavior data is classified as abnormal behavior data;

[0022] If the camouflage coefficient is less than a preset camouflage coefficient reference threshold, the acquired behavior data is classified as normal behavior data.

[0023] Preferably, when abnormal behavior data is generated after analyzing the acquired user behavior data under a fixed detection window, a plurality of camouflage coefficients generated after analyzing the user behavior data under the fixed detection window are continuously acquired to establish an analysis set, and the camouflage coefficients in the analysis set are compared with the first-level reference threshold, the second-level reference threshold, and the camouflage coefficient reference threshold, wherein the second-level reference threshold is greater than the first-level reference threshold, and the first-level reference threshold is greater than the camouflage coefficient reference threshold, and the camouflage coefficients are compared and analyzed with the second-level reference threshold, the first-level reference threshold, and the camouflage coefficient reference threshold, and the number of camouflage coefficients that are less than the first-level reference threshold and greater than or equal to the camouflage coefficient reference threshold is calibrated as Fa, the number of camouflage coefficients that are less than the second-level reference threshold and greater than or equal to the first-level reference threshold is calibrated as Fb, and the number of camouflage coefficients that are greater than or equal to the second-level reference threshold is calibrated as Fc;

[0024] Comprehensively analyze Fa, Fb, and Fc to generate the abnormal risk factor ARI, based on the formula:

[0025]

[0026] , where k1, k2, k3 are the preset proportional coefficients of Fa, Fb, Fc respectively, and k1, k2, k3 are all greater than 0.

[0027] Preferably, the abnormal risk coefficient generated after further comprehensive analysis after the acquired behavior data is divided into abnormal behavior data is compared and analyzed with a preset first abnormal risk coefficient reference threshold and a preset second abnormal risk coefficient reference threshold to determine the abnormal risk level of the acquired behavior data. The comparative analysis results are as follows:

[0028] If the abnormal risk coefficient is less than the preset first abnormal risk coefficient reference threshold, the acquired behavior data abnormal risk level is classified as low risk abnormality;

[0029] If the abnormal risk coefficient is greater than or equal to a preset first abnormal risk coefficient reference threshold and less than a preset second abnormal risk coefficient reference threshold, the acquired behavior data abnormal risk level is classified as medium risk abnormal;

[0030] If the abnormal risk coefficient is greater than or equal to a preset second abnormal risk coefficient reference threshold, the acquired behavior data abnormal risk level is classified as a high-risk abnormality.

[0031] Preferably, in the detection window, the contrast between the user's click, browsing behavior and purchase behavior is analyzed, and the specific steps of generating the purchase conversion contradiction index are as follows:

[0032] In the detection window, the weights of clicks and browsing are defined according to the number of clicks and browsing time of each product by the user. The calculation expression is as follows:

[0033]

[0034] , where C w is the click and browse weight, c i is the number of clicks on product i by the user, t i is the browsing time of the user on product i, and n is the total number of browsed products;

[0035] Define the user's purchase behavior index. The purchase behavior index combines the user's actual purchase behavior and the amount of money the user spends on each product. The calculation expression is as follows:

[0036]

[0037] , where P a is the purchasing behavior index, p i is the purchase frequency of product i by the user, a i is the single purchase amount of the user for product i;

[0038] Based on click-through weight C w and purchasing behavior index P a , define the degree of behavioral contrast, which is an exponential transformation of the ratio of the two to amplify inconsistent behavioral performance. The calculation expression is as follows:

[0039]

[0040] , where D is the degree of behavioral contrast and e is the natural base;

[0041] The behavior contrast degree D is converted into the purchase conversion contradiction index, and logarithmic smoothing is used to control data fluctuations. The calculation expression is as follows:

[0042] PCCI=ln(1+D 2 )

[0043] , where PCCI is the purchase conversion contradiction index.

[0044] Preferably, under the detection window, the correlation between user behavior and the platform reward mechanism is analyzed, and the specific steps of generating the task-oriented index are as follows:

[0045] In the detection window, the number of behavioral events triggered by the user during the reward activity is recorded to reflect the behavior pattern of the user performing a large number of tasks in a short period of time. The density of the user-triggered behavioral events is calculated by the number of behavioral events triggered by the user during the reward activity. The calculation expression is as follows:

[0046]

[0047] , where M is the user behavior trigger density, f k is the number of task triggering events of the user in the kth period during the reward activity, w k is the task weight of the kth period during the reward activity, e is the natural base number, and m is the total number of periods during the reward activity;

[0048] Calculate the user behavior reward bias. The user behavior reward bias is used to reflect the degree of bias of the user in the reward task. The calculation expression is as follows:

[0049]

[0050] , where B is the user behavior reward bias, r l is the completion rate of the user in the lth task, ω l is the reward weight of the lth task, which determines the degree of influence of the task on the overall behavior reward bias, and d is the number of task types in which the user participates;

[0051] Calculate the user behavior repetition tendency. The user behavior repetition tendency is used to detect whether the user repeatedly performs the same task in the reward activity. The calculation expression is as follows:

[0052]

[0053] , where R is the user behavior repetition tendency, q j is the number of times the user repeatedly triggers the jth task, β j is the weight of repeated behavior, and p is the number of task types triggered by the user during the activity;

[0054] The task-oriented index is generated by combining the user behavior trigger density M, the user behavior reward bias B, and the user behavior repetition tendency R. The calculation expression is as follows:

[0055]

[0056] , where TOI is the task orientation index, γ is the amplification factor, δ is an adjustment factor, and ∈ is a small constant used to prevent the denominator from being zero.

[0057] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0058] Through multi-source data collection and intelligent feature extraction and evaluation modules, the system of the present invention can more accurately identify the real needs of users and avoid misleading gift design by disguised behavior. Disguised behavior is identified by the purchase conversion contradiction index and the task-oriented index, and the user's natural behavior is separated from the abnormal behavior, which helps to generate a more practical gift design plan. Furthermore, by performing risk assessment on abnormal behavior through the risk level analysis and response module, the system can take effective measures to filter high-risk data, thereby ensuring that the final design plan meets the user's real interests and needs. This not only improves the accuracy of gift design, but also improves user satisfaction and the attractiveness of the event, thereby enhancing the brand image and optimizing the user's overall experience of the event.

[0059] The present invention divides normal and abnormal behavior data through a data classification module, and the system can effectively reduce the negative impact of disguised behavior on the data model, thereby improving the long-term stability of the model. In particular, with the support of abnormal risk index assessment, the system can dynamically adjust the way abnormal data is handled and allocate different resources and measures according to the risk level. This flexible processing mechanism not only ensures data quality, but also avoids the waste of resources caused by excessive cleaning or screening. At the same time, the system can use the processed data in a more efficient way in subsequent analysis, reducing the impact of disguised behavior on model iteration, making the overall system more efficient and accurate when dealing with complex data environments, laying a solid foundation for future theme activity design. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0061] Figure 1 The present invention is a module schematic diagram of a promotional gift customization design system based on theme activities. DETAILED DESCRIPTION

[0062] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0063] The present invention provides Figure 1The promotional gift customization design system based on theme activities shown in the figure includes a multi-source data acquisition module, an intelligent feature extraction and evaluation module, a data classification module, and a risk level analysis and response module:

[0064] The multi-source data collection module first conducts comprehensive and multi-source data collection to obtain user behavior data on various platforms and channels. By integrating multi-dimensional data, it builds a comprehensive portrait of the user, laying a solid foundation for subsequent analysis.

[0065] The steps to conduct comprehensive and multi-source data collection, build a comprehensive portrait of the user, and lay a solid foundation for subsequent analysis are as follows:

[0066] First, identify all data sources and determine the scope of user behavior data to be collected;

[0067] This includes online channels (such as websites, mobile applications, and social media platforms) and offline channels (such as store purchase records and offline event participation). It can also be expanded to third-party data sources, such as advertising platforms and user feedback systems, to ensure that user behavior in various channels is covered. Clarifying the scope and source of data collection helps ensure the comprehensiveness of the data and provides rich basic information for a comprehensive user portrait.

[0068] After determining the data source, set the corresponding data collection frequency and sampling strategy;

[0069] Some data needs to be collected in real time, such as the user's click stream on the website, while other data can be collected on a daily or weekly basis, such as purchase records and visit frequency. Determine a reasonable sampling strategy based on the characteristics of the data and avoid excessive collection, which can not only ensure the real-time nature of the data, but also reduce unnecessary resource consumption. At the same time, balance the frequency and coverage of collection so as to control the quality and quantity of data while obtaining accurate user behavior data.

[0070] Preprocess the collected data;

[0071] By removing invalid data, filling missing values, converting formats, and other processing steps, data from different sources are made to have consistent formats and standards. At the same time, unstructured data (such as text comments or multimedia files) are processed as necessary and converted into structured data or feature information so that these data can be effectively used in subsequent analysis to improve the integrity of user portraits.

[0072] After cleaning and standardization, data from different platforms and channels are integrated to form a comprehensive profile of the user;

[0073] During the integration process, ensure that data from different sources can accurately match the same user through unique identification. Based on the integrated data, build a multi-dimensional portrait of the user, including behavioral characteristics (such as visit frequency, preferred categories), demographic information (such as age, gender), interests and hobbies, etc. This step ensures that the system can understand user behavior and preferences from multiple perspectives, providing a high-quality foundation for subsequent analysis and personalized recommendations.

[0074] The intelligent feature extraction and evaluation module extracts features from the acquired behavior data, inputs the extracted feature information into the pre-learned machine learning model, and intelligently evaluates the acquired user behavior data to predict the authenticity and preference of user behavior and identify disguised behavior in the behavior data;

[0075] Extract features from the acquired behavior data. The extracted features include the contrast between the user's click, browsing behavior and purchase behavior and the correlation between the user's behavior and the platform reward mechanism. After analyzing the contrast between the extracted user's click, browsing behavior and purchase behavior, a purchase conversion contradiction index is generated. After analyzing the correlation between the extracted user behavior and the platform reward mechanism, a task orientation index is generated. The disguised behavior in the behavior data is identified through the purchase conversion contradiction index and the task orientation index.

[0076] The abnormal contrast between the user's click, browsing behavior and purchase behavior usually indicates that the obtained user behavior data may be abnormal data generated by disguised behavior. Real user behavior usually shows a certain conversion trend, that is, after browsing or clicking a certain number of products, they will gradually generate behaviors such as purchase, adding to shopping carts or further consultation. Disguised behavior is often limited to a large number of clicks and browsing operations, but does not show real purchasing interest or behavior conversion. This abnormal contrast data feature may be a disguised behavior generated by users in order to obtain click or browsing rewards from the platform, rather than out of actual consumption needs. In addition, disguised behavior often focuses on high-frequency clicks in a short period of time, lacks continuity and correlation, making it difficult for the system to find a reasonable behavior logic, further exacerbating the contradiction between click, browse and purchase behaviors. Therefore, this abnormal contrast phenomenon is an effective indicator for identifying disguised behavior in data analysis.

[0077] In the detection window, the contrast between the user's click, browsing behavior and purchase behavior is analyzed, and the specific steps for generating the purchase conversion contradiction index are as follows:

[0078] In the detection window, the weights of clicks and browsing are defined according to the number of clicks and browsing time of each product by the user. The calculation expression is as follows:

[0079]

[0080] , where C w is the click and browse weight, which is the comprehensive weight calculated by the user's click and browse behavior on each product within the detection window, and is used to quantify the user's "stay degree" on multiple product pages. i is the number of clicks on product i by the user, t i is the browsing time of the user on product i, and n is the total number of browsed products;

[0081] The click and browse weights represent the overall intensity of the user's behavior on multiple product pages within the detection window. By measuring the user's click frequency and dwell time on each product, the user's interest and engagement are quantified. Here, its role is to convert the user's click and browse behavior into a quantifiable index for comparison with purchase behavior. When the click and browse weights are significantly higher than the purchase behavior, it may indicate that the user is just pretending to complete a task (such as earning points), thereby helping the system identify abnormal data and potential disguised behavior.

[0082] Define the user's purchase behavior index. The purchase behavior index combines the user's actual purchase behavior and the amount of money the user spends on each product. The calculation expression is as follows:

[0083]

[0084] , where P a is the purchase behavior index, which measures the purchase behavior intensity of users within the detection window, p i is the purchase frequency of product i by the user, a i is the single purchase amount of the user for product i;

[0085] Based on click-through weight C w and purchasing behavior index P a , define the degree of behavioral contrast, which is an exponential transformation of the ratio of the two to amplify inconsistent behavioral performance. The calculation expression is as follows:

[0086]

[0087] , where D is the degree of behavioral contrast and e is the natural base;

[0088] If you click to browse weight C w Much greater than the purchasing behavior index P a , then the behavior contrast degree D will show exponential growth, indicating that click-through behavior far exceeds purchase behavior and is abnormal data; and if the click-through weight C w Purchase behavior index P a If they are relatively close, the behavior contrast degree D value is relatively low, which is a normal user behavior.

[0089] The behavior contrast degree D is converted into the purchase conversion contradiction index, and logarithmic smoothing is used to control data fluctuations. The calculation expression is as follows:

[0090] PCCI=ln(1+D 2 )

[0091] , where PCCI is the purchase conversion contradiction index;

[0092] The process of converting the behavioral contrast degree D into the purchase conversion contradiction index PCCI is realized through a logarithmic smoothing formula to more stably measure the difference between click-through behavior and purchase behavior. The conversion formula is PCCI = ln(1+D 2 ), where D is amplified by the square, which can further highlight the greater contrast, making it easier to identify high-contrast abnormal behaviors as disguised behaviors. At the same time, the definition of this index smoothes the range of the original behavior contrast degree D to a standardized scale, making the value of the purchase conversion contradiction index PCCI comparable among different users.

[0093] Logarithmic smoothing is used to control data fluctuations, especially for abnormally high behavioral contrast levels D. Logarithmic transformation limits the growth rate of extreme values ​​to a lower level, avoiding the phenomenon of index imbalance caused by individual extreme high values. In addition, logarithmic smoothing helps reduce the impact of noisy data, making the purchase conversion contradiction index PCCI more robust to small behavioral changes. In this way, when processing user behavior data, the system can more reliably identify real abnormal behaviors, thereby accurately identifying disguised behaviors in large-scale user data.

[0094] Under the detection window, the larger the purchase conversion contradiction index performance value generated by analyzing the contrast between the user's click, browsing behavior and purchase behavior, the more likely it is that the obtained user behavior data is abnormal data generated by disguised behavior. This index determines the authenticity of the data by analyzing the contrast between the user's click and browsing behavior and the actual purchase behavior. When users click and browse a large number of products but have almost no actual purchase or shopping cart behavior, the index value will increase, which is usually a typical feature of disguised behavior, because real users generally have a certain probability of purchasing after browsing. On the contrary, if the purchase conversion contradiction index is low, it indicates that there is a certain conversion trend between click, browse and purchase behaviors, which is in line with normal user behavior patterns.

[0095] When the correlation between user behavior and the platform reward mechanism is highly consistent, this usually indicates that the user behavior data may be abnormal data generated by disguised behavior. A high correlation means that the user's actions are significantly focused on obtaining rewards rather than expressing real interests or needs. For example, users frequently click on specific products during a points or reward event, but the behavior quickly decreases or even stops after the event. This behavioral fluctuation is closely linked to the reward mechanism, showing that users are more concerned about the reward itself rather than the product. Disguised behavior usually aims to complete tasks and reach reward thresholds, showing overly consistent or short-term behavioral patterns, lacking randomness and natural consumption paths. Therefore, when the user's actions are highly consistent with the platform reward mechanism, and the behavior shows extreme or unnatural fluctuations, it often means that these data are not a reflection of real needs, but abnormal data driven by rewards.

[0096] Under the detection window, the correlation between user behavior and the platform reward mechanism is analyzed, and the specific steps for generating the task-oriented index are as follows:

[0097] In the detection window, the number of behavioral events triggered by the user during the reward activity is recorded to reflect the behavior pattern of the user performing a large number of tasks in a short period of time. The density of the user-triggered behavioral events is calculated by the number of behavioral events triggered by the user during the reward activity. The calculation expression is as follows:

[0098]

[0099] , where M is the user behavior trigger density, f k is the number of task triggering events of the user in the kth period during the reward activity, w k is the task weight of the kth period during the reward activity, e is the natural base number, and m is the total number of periods during the reward activity;

[0100] The purpose of user behavior trigger density is to measure how frequently users perform tasks during reward activities to identify whether there is abnormal operation density. By calculating the number of task events triggered by users in a short period of time and the weights of these events in different time periods, behavior trigger density can reveal whether users exhibit high-frequency operation patterns beyond the norm. This high-density task triggering is usually to quickly accumulate points or complete task goals, with obvious purpose and lack of natural behavior rhythm. Therefore, behavior trigger density can effectively help the system identify potential disguised behaviors and provide an important reference dimension for anomaly detection.

[0101] Calculate the user behavior reward bias. The user behavior reward bias is used to reflect the degree of bias of the user in the reward task. The calculation expression is as follows:

[0102]

[0103] , where B is the user behavior reward bias, r l is the completion ratio of the user in the lth task, that is, the ratio of the number of times a user completes a specific task to the number of times a user completes other tasks, ω l is the reward weight of the lth task, which determines the degree of influence of the task on the overall behavior reward bias, and d is the number of task types in which the user participates;

[0104] The role of user behavior reward bias is to measure the user's preference for specific tasks in the reward activity, thereby revealing whether the user intentionally focuses on certain tasks to obtain rewards. For example, if the user is highly focused on the "check-in" and "sharing" tasks during the reward activity and ignores other tasks, this indicates that the user may be more concerned about the reward returns of these tasks and may have a task-oriented behavior pattern. By analyzing the user's participation and bias in different tasks, their operating motivations can be better identified and the authenticity and naturalness of the behavior can be judged.

[0105] Calculate the user behavior repetition tendency. The user behavior repetition tendency is used to detect whether the user repeatedly performs the same task in the reward activity. The calculation expression is as follows:

[0106]

[0107] , where R is the user behavior repetition tendency, q j is the number of times the user repeatedly triggers the jth task, β j is the weight of repeated behavior, and p is the number of task types triggered by the user during the activity;

[0108] There is a subtle but important difference between the number of tasks a user participates in and the number of tasks a user triggers during an activity:

[0109] The number of task types a user participates in refers to the total number of task types that the user actively selects and participates in during the reward activity. This indicates which different task types the user intentionally joins, regardless of how frequently each task is triggered. For example, if the activity has tasks such as "check in", "share", "browse products", and "place an order", and the user only chooses to participate in the "check in" and "share" tasks, then the number of task types the user participates in is 2.

[0110] The number of task types triggered by users during an activity refers to the number of task types that users actually perform and generate records for. This includes all tasks that are recorded during the activity, regardless of whether they are actively selected by the user. For example, a user performs the three tasks of "check-in", "sharing", and "browsing products" during an activity. Although the "browsing products" task is not one of the tasks that the user initially actively selected, it is still counted as a triggered task type because it generates a trigger record, which is 3.

[0111] Therefore, the number of participated task types reflects the range of tasks actively selected by users, while the number of triggered task types more comprehensively includes all operation records of users during the activity, which may cover some tasks that users did not explicitly select but actually performed.

[0112] The purpose of user behavior repetition tendency is to evaluate whether there is a behavioral pattern of users repeatedly performing the same task in reward activities to identify possible abnormal operations. Frequent repetition of the same task is usually not a natural behavior, but a disguised behavior driven by rewards. For example, if a user performs the "check-in" task multiple times in a short period of time, it may be to quickly accumulate points rather than out of real interest or need. Behavior repetition tendency can effectively help the system identify these too frequent task triggering behaviors, prompting the user's operation may be abnormal, thereby improving the detection accuracy of disguised behaviors.

[0113] The task-oriented index is generated by combining the user behavior trigger density M, the user behavior reward bias B, and the user behavior repetition tendency R. The calculation expression is as follows:

[0114]

[0115] , where TOI is the task orientation index, γ is the amplification factor, which controls the weight of the square term on the overall index, δ is an adjustment coefficient, which is used to control the influence of the square root interaction term to avoid excessive fluctuations at high values, and ∈ is a small constant used to prevent the denominator from being zero;

[0116] The amplification factor γ is a coefficient used to control the influence of a certain formula item in the overall calculation. In the task-oriented index formula, γ controls the weight of the square term correction. Specifically, it determines the amplification effect of the behavior trigger density M under special circumstances. Since the square term will significantly increase the impact of the behavior trigger density M, if it is not properly adjusted, it may cause the formula to be too sensitive to high-frequency behavior. Therefore, the contribution of the square term to the task-oriented index can be flexibly adjusted through the amplification factor γ, so that the term can amplify frequently triggered behavior signals without causing extreme values ​​to get out of control, thereby improving the accuracy of identifying disguised behaviors.

[0117] The adjustment coefficient δ is a coefficient used to adjust the degree of influence of the interaction term between two variables. In the formula, δ controls the weight of the square root interaction term of the behavior reward bias B and the behavior repetition tendency R. This interaction term is designed to capture the behavioral characteristics of users that exhibit both reward orientation and repetitiveness. Through the adjustment coefficient δ, the impact of this term on the overall task orientation index can be dynamically adjusted, so that when both the behavior reward bias B and the behavior repetition tendency R are high, the effect of this term is appropriately amplified, but it will not have an overly strong impact under normal circumstances. This adjustment mechanism makes the effect of the interaction term softer, thereby avoiding the excessive influence of a single behavioral characteristic on the task orientation judgment.

[0118] Under the detection window, the larger the performance value of the task-oriented index generated after analyzing the correlation between user behavior and the platform reward mechanism, the more it usually indicates that the user behavior data is abnormal data generated by disguised behavior. Specifically, a higher task-oriented index means that the user's behavior pattern is highly concentrated on the triggering conditions of the points or reward activities, such as frequently performing tasks, clicking, browsing, and other operations in a short period of time to achieve the reward goal, and the behavior is significantly reduced or stopped after the activity ends. Such highly correlated operations usually lack the naturalness and diversity of normal user behavior, showing their purposefulness and artificial characteristics. On the contrary, if the task-oriented index is low, it means that the correlation between user behavior and the platform reward mechanism is weak, and it is more likely to reflect real interests and needs, which is normal behavior data.

[0119] After obtaining the purchase conversion contradiction index and the task orientation index, the purchase conversion contradiction index PCCI and the task orientation index TOI are input into the pre-learned machine learning model, and the disguise coefficient CC is generated by the machine learning model, and the obtained behavior data is intelligently evaluated by the disguise coefficient CC;

[0120] The machine learning model is not limited here, and any machine learning model that can generate a disguise coefficient CC after comprehensive analysis of the purchase conversion contradiction index PCCI and the task orientation index TOI can be used. In order to implement the technical solution of the present invention, the present invention provides a specific implementation method:

[0121] The calculation formula for the camouflage coefficient CC is as follows:

[0122]

[0123] , where f1 and f2 are the preset proportional coefficients of the conversion contradiction index PCCI and the task orientation index TOI, respectively, and both f1 and f2 are greater than 0.

[0124] It can be seen from the calculation expression of the camouflage coefficient that, under the detection window, the larger the purchase conversion contradiction index performance value generated after analyzing the contrast between the user's click, browsing behavior and purchasing behavior, the larger the task-oriented index performance value generated after analyzing the correlation between the user behavior and the platform reward mechanism, indicating that under the detection window, the larger the camouflage coefficient performance value generated after analyzing the acquired user behavior data, the greater the abnormal behavior data generated by camouflage behavior, and vice versa.

[0125] The data classification module divides the acquired behavior data into normal behavior data and abnormal behavior data based on the evaluation results of the machine learning model;

[0126] The camouflage coefficient generated after analyzing the acquired user behavior data is compared and analyzed with the preset camouflage coefficient reference threshold, and the acquired behavior data is divided. The division steps are as follows:

[0127] If the camouflage coefficient is greater than or equal to a preset camouflage coefficient reference threshold, the acquired behavior data is classified as abnormal behavior data;

[0128] If the camouflage coefficient is less than a preset camouflage coefficient reference threshold, the acquired behavior data is classified as normal behavior data;

[0129] Normal behavior data refers to the natural and real operation records of users on the platform, which reflects the users' actual needs and interests. Abnormal behavior data refers to the operation records of users that are driven by specific factors (such as point rewards, task requirements, etc.) and do not conform to normal logic, which often do not reflect their real needs.

[0130] The risk level analysis and response module conducts further comprehensive analysis on the identified abnormal behavior data, determines the risk level of the abnormal behavior data, and takes different response measures for different risk levels;

[0131] When abnormal behavior data is generated after analyzing the acquired user behavior data under a fixed detection window, a number of camouflage coefficients generated after analyzing the user behavior data under the fixed detection window are continuously acquired to establish an analysis set, and the camouflage coefficients in the analysis set are compared with a first-level reference threshold, a second-level reference threshold, and a camouflage coefficient reference threshold, wherein the second-level reference threshold is greater than the first-level reference threshold, and the first-level reference threshold is greater than the camouflage coefficient reference threshold, and the camouflage coefficients are compared and analyzed with the second-level reference threshold, the first-level reference threshold, and the camouflage coefficient reference threshold, and the number of camouflage coefficients that are less than the first-level reference threshold and greater than or equal to the camouflage coefficient reference threshold is calibrated as Fa, the number of camouflage coefficients that are less than the second-level reference threshold and greater than or equal to the first-level reference threshold is calibrated as Fb, and the number of camouflage coefficients that are greater than or equal to the second-level reference threshold is calibrated as Fc;

[0132] Comprehensively analyze Fa, Fb, and Fc to generate the abnormal risk factor ARI, based on the formula:

[0133]

[0134] , where k1, k2, k3 are the preset proportional coefficients of Fa, Fb, Fc respectively, and k1, k2, k3 are all greater than 0.

[0135] It can be seen from the calculation expression of the abnormal risk coefficient that the larger the value of the abnormal risk coefficient is, the higher the risk brought by the abnormal behavior data is, and vice versa.

[0136] The acquired behavior data is divided into abnormal behavior data, and then the abnormal risk coefficient generated after further comprehensive analysis is compared and analyzed with the preset first abnormal risk coefficient reference threshold and the preset second abnormal risk coefficient reference threshold to determine the abnormal risk level of the acquired behavior data. The comparative analysis results are as follows:

[0137] If the abnormal risk coefficient is less than the preset first abnormal risk coefficient reference threshold, the acquired behavior data abnormal risk level is classified as low risk abnormality;

[0138] If the abnormal risk coefficient is greater than or equal to a preset first abnormal risk coefficient reference threshold and less than a preset second abnormal risk coefficient reference threshold, the acquired behavior data abnormal risk level is classified as medium risk abnormal;

[0139] If the abnormal risk coefficient is greater than or equal to a preset second abnormal risk coefficient reference threshold, the acquired behavior data abnormal risk level is classified as a high-risk abnormality.

[0140] Low-risk exception response measures

[0141] For low-risk abnormal data, since their risk level is low and their impact on the overall system is limited, relatively relaxed countermeasures can be taken. For example, these data can be marked as potential abnormal data, but they are not immediately excluded from the analysis process and are retained in the data pool for subsequent observation. At the same time, a lightweight data correction strategy can be applied to low-risk abnormal data to perform partial weight reduction processing, so that their weight in the algorithm model is slightly reduced to reduce their interference with the overall analysis results. This flexible processing method does not affect the weight of normal data, and leaves room for observation for possible subsequent abnormal increases.

[0142] Medium-risk abnormal response measures

[0143] For medium-risk abnormal data, more stringent processing measures should be taken. This type of data indicates that user behavior may have a tendency to disguise, and the impact on system analysis and design results is relatively significant. This data can be included in the key monitoring objects and reviewed regularly, and its risk changes can be screened and judged using automated rules. At the same time, medium-risk abnormal data can be isolated from the main behavioral data, stored separately and subjected to detailed feature analysis to explore its possible behavior patterns and causes. The weight of this type of data should be significantly reduced, and it can even be set to directly exclude it from the data analysis and recommendation system when the threshold reaches a certain level to avoid its interference with the algorithm.

[0144] High-risk abnormal response measures

[0145] For high-risk abnormal data, decisive exclusion measures should be taken because of its high possibility of disguise and high risk level. Such data can easily cause deviations to the analysis system and recommendation algorithm, so it can be immediately removed from the data set to prevent it from affecting system decisions. At the same time, a data isolation and ban mechanism for high-risk abnormal behaviors should be established, and users who have repeatedly exhibited such behaviors should be specifically marked, or their accounts should be restricted according to platform policies. Detailed behavior reports can be further generated, and high-risk abnormal data can be archived and analyzed regularly to identify possible disguised behavior patterns and provide data support for future disguise detection and prevention strategies.

[0146] Through multi-source data collection and intelligent feature extraction and evaluation modules, the system of the present invention can more accurately identify the real needs of users and avoid misleading gift design by disguised behavior. Disguised behavior is identified by the purchase conversion contradiction index and the task-oriented index, and the user's natural behavior is separated from the abnormal behavior, which helps to generate a more practical gift design plan. Furthermore, by performing risk assessment on abnormal behavior through the risk level analysis and response module, the system can take effective measures to filter high-risk data, thereby ensuring that the final design plan meets the user's real interests and needs. This not only improves the accuracy of gift design, but also improves user satisfaction and the attractiveness of the event, thereby enhancing the brand image and optimizing the user's overall experience of the event.

[0147] The present invention divides normal and abnormal behavior data through a data classification module, and the system can effectively reduce the negative impact of disguised behavior on the data model, thereby improving the long-term stability of the model. In particular, with the support of abnormal risk index assessment, the system can dynamically adjust the way abnormal data is handled and allocate different resources and measures according to the risk level. This flexible processing mechanism not only ensures data quality, but also avoids the waste of resources caused by excessive cleaning or screening. At the same time, the system can use the processed data in a more efficient way in subsequent analysis, reducing the impact of disguised behavior on model iteration, making the overall system more efficient and accurate when dealing with complex data environments, laying a solid foundation for future theme activity design.

[0148] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A promotional gift customization design system based on theme activities, characterized in that: It includes multi-source data acquisition module, intelligent feature extraction and evaluation module, data classification module and risk level analysis and response module: The multi-source data collection module first conducts comprehensive and multi-source data collection to obtain user behavior data on various platforms and channels. By integrating multi-dimensional data, it builds a comprehensive portrait of the user, laying a solid foundation for subsequent analysis. The intelligent feature extraction and evaluation module extracts features from the acquired behavior data, inputs the extracted feature information into the pre-learned machine learning model, and intelligently evaluates the acquired user behavior data to predict the authenticity and preference of user behavior and identify disguised behavior in the behavior data; The data classification module divides the acquired behavior data into normal behavior data and abnormal behavior data based on the evaluation results of the machine learning model; The risk level analysis and response module conducts further comprehensive analysis on the identified abnormal behavior data, determines the risk level of the abnormal behavior data, and takes different response measures for different risk levels.

2. A promotional gift customization design system based on theme activities according to claim 1, characterized in that: The steps to conduct comprehensive and multi-source data collection, build a comprehensive portrait of the user, and lay a solid foundation for subsequent analysis are as follows: First, identify all data sources and determine the scope of user behavior data to be collected; After determining the data source, set the corresponding data collection frequency and sampling strategy; Preprocess the collected data; After cleaning and standardization, data from different platforms and channels are integrated to form a comprehensive portrait of the user.

3. A promotional gift customization design system based on theme activities according to claim 1, characterized in that: Feature extraction is performed on the acquired behavioral data. The extracted features include the contrast between the user's click, browsing behavior and purchasing behavior and the correlation between the user behavior and the platform reward mechanism. After analyzing the contrast between the extracted user's click, browsing behavior and purchasing behavior, a purchase conversion contradiction index is generated. After analyzing the correlation between the extracted user behavior and the platform reward mechanism, a task-oriented index is generated. The purchase conversion contradiction index and the task-oriented index are used to identify disguised behaviors in the behavioral data.

4. A promotional gift customization design system based on theme activities according to claim 3, characterized in that: After obtaining the purchase conversion contradiction index and the task orientation index, the purchase conversion contradiction index and the task orientation index are input into the pre-learned machine learning model, and the camouflage coefficient is generated by the machine learning model. The camouflage coefficient is used to perform an intelligent evaluation of the obtained behavioral data.

5. A promotional gift customization design system based on theme activities according to claim 4, characterized in that: The camouflage coefficient generated after analyzing the acquired user behavior data is compared and analyzed with the preset camouflage coefficient reference threshold, and the acquired behavior data is divided. The division steps are as follows: If the camouflage coefficient is greater than or equal to a preset camouflage coefficient reference threshold, the acquired behavior data is classified as abnormal behavior data; If the camouflage coefficient is less than a preset camouflage coefficient reference threshold, the acquired behavior data is classified as normal behavior data.

6. A promotional gift customization design system based on theme activities according to claim 5, characterized in that: When abnormal behavior data is generated after analyzing the acquired user behavior data under a fixed detection window, a number of camouflage coefficients generated after analyzing the user behavior data under the fixed detection window are continuously acquired to establish an analysis set, and the camouflage coefficients in the analysis set are compared with the first-level reference threshold, the second-level reference threshold, and the camouflage coefficient reference threshold, wherein the second-level reference threshold is greater than the first-level reference threshold, and the first-level reference threshold is greater than the camouflage coefficient reference threshold, and the camouflage coefficient is compared and analyzed with the second-level reference threshold, the first-level reference threshold, and the camouflage coefficient reference threshold, and the number of camouflage coefficients that are less than the first-level reference threshold and greater than or equal to the camouflage coefficient reference threshold is calibrated as Fa, the number of camouflage coefficients that are less than the second-level reference threshold and greater than or equal to the first-level reference threshold is calibrated as Fb, and the number of camouflage coefficients that are greater than or equal to the second-level reference threshold is calibrated as F c ; Comprehensively analyze Fa, Fb, and Fc to generate the abnormal risk factor ARI, based on the formula: , where k1, k2, k3 are the preset proportional coefficients of Fa, Fb, Fc respectively, and k1, k2, k3 are all greater than 0.

7. A promotional gift customization design system based on theme activities according to claim 6, characterized in that: The acquired behavior data is divided into abnormal behavior data, and then the abnormal risk coefficient generated after further comprehensive analysis is compared and analyzed with the preset first abnormal risk coefficient reference threshold and the preset second abnormal risk coefficient reference threshold to determine the abnormal risk level of the acquired behavior data. The comparative analysis results are as follows: If the abnormal risk coefficient is less than the preset first abnormal risk coefficient reference threshold, the acquired behavior data abnormal risk level is classified as low risk abnormality; If the abnormal risk coefficient is greater than or equal to a preset first abnormal risk coefficient reference threshold and less than a preset second abnormal risk coefficient reference threshold, the acquired behavior data abnormal risk level is classified as medium risk abnormal; If the abnormal risk coefficient is greater than or equal to a preset second abnormal risk coefficient reference threshold, the acquired behavior data abnormal risk level is classified as a high-risk abnormality.

8. The promotional gift customization design system based on theme activities according to claim 3 is characterized in that: In the detection window, the contrast between the user's click, browsing behavior and purchase behavior is analyzed, and the specific steps for generating the purchase conversion contradiction index are as follows: In the detection window, the weights of clicks and browsing are defined according to the number of clicks and browsing time of each product by the user. The calculation expression is as follows: , where C w is the click and browse weight, c i is the number of clicks on product i by the user, t i is the browsing time of the user on product i, and n is the total number of browsed products; Define the user's purchase behavior index. The purchase behavior index combines the user's actual purchase behavior and the amount of money the user spends on each product. The calculation expression is as follows: , where P a is the purchasing behavior index, p i is the purchase frequency of product i by the user, a i is the single purchase amount of the user for product i; Based on click-through weight C w and purchasing behavior index P a , define the degree of behavioral contrast, which is an exponential transformation of the ratio of the two to amplify inconsistent behavioral performance. The calculation expression is as follows: , where D is the degree of behavioral contrast and e is the natural base; The behavior contrast degree D is converted into the purchase conversion contradiction index, and logarithmic smoothing is used to control data fluctuations. The calculation expression is as follows: PCCI=ln(1+D 2 ), where PCCI is the purchase conversion contradiction index.

9. The promotional gift customization design system based on theme activities according to claim 3 is characterized in that: Under the detection window, the correlation between user behavior and the platform reward mechanism is analyzed, and the specific steps for generating the task-oriented index are as follows: In the detection window, the number of behavioral events triggered by the user during the reward activity is recorded to reflect the behavior pattern of the user performing a large number of tasks in a short period of time. The density of the user-triggered behavioral events is calculated by the number of behavioral events triggered by the user during the reward activity. The calculation expression is as follows: , where M is the user behavior trigger density, fk is the number of task trigger events of the user in the kth period during the reward activity, and w k is the task weight of the kth period during the reward activity, e is the natural base number, and m is the total number of periods during the reward activity; Calculate the user behavior reward bias. The user behavior reward bias is used to reflect the degree of bias of the user in the reward task. The calculation expression is as follows: , where B is the user behavior reward bias, rl is the user's completion ratio in the lth task, ω l is the reward weight of the lth task, which determines the degree of influence of the task on the overall behavior reward bias, and d is the number of task types in which the user participates; Calculate the user behavior repetition tendency. The user behavior repetition tendency is used to detect whether the user repeatedly performs the same task in the reward activity. The calculation expression is as follows: , where R is the user behavior repetition tendency, q j is the number of times the user repeatedly triggers the jth task, β j is the weight of repeated behavior, and p is the number of task types triggered by the user during the activity; The task-oriented index is generated by combining the user behavior trigger density M, the user behavior reward bias S, and the user behavior repetition tendency R. The calculation expression is as follows: , where TOI is the task orientation index, γ is the amplification factor, δ is an adjustment factor, and ∈ is a small constant used to prevent the denominator from being zero.

Images